Hacker News Reader

Comment by maelito | Hacker News Reader

We absolutely need certified no-AI digital proofs.

We absolutely also need full and complete peace on earth and no more wars, but what we need isn't always what is feasible to get in real life. "No-AI proofs" falls into that bucket of "Would be nice to have, but very infeasible to actually create".

it will be implemented by chips-on-camera, that will tie you to a picture. it will be pushed forward by law enforcement ("think of the children!") and it will be great for profits. not so much for privacy.

Yet it doesn't actually solve the problem. Whatever technical implementation you can think of, can somehow be misused so you have AI pictures labeled as "Genuinely not AI".

If you do come up with a 100% fool-proof implementation of this, you'll be able to get a lot of money for it, so do give it a try! Many have tried before you, yet it always turns out to be short of impossible. But who knows, maybe there is a way...

I didn't say it would work - I said it would be made mandatory and will be used against us.

Right, I guess I was still with the original topic/subject:

> We absolutely need certified no-AI digital proofs.

Out of curiosity, what stops you from taking a photo of a AI generated picture?

Well, you could make your verified camera do a 3d scan. Then at lest you'd need AI to 3d print a scene or something.

How do you securely read this "3D scan" sensor data, being 100% it hasn't been tampered with?

Use whatever technology you used to make the 2d camera tamperproof.

you can tie the sensor to a chip that signs the data as it goes out.

The same thing that stops your phone’s Face ID working with a photo of a face, I suppose

>it will be implemented by chips-on-camera, that will tie you to a picture.

You can mitigate this by having a pool of devices (eg. 1000) share keys. AFAIK TPM chips and U2F/FIDO keys do this to provide some anonymity while limiting the blast radius if a key does get leaked.

Well that's no good for big brother, is it.

This seems like such a tricky problem. There was a no-ai camera posted here recently which verified the photos were genuinely taken on the device. It was pointed out someone could photograph an ai image via the camera to produce a verified image.

Maybe it needs to be similar to SSL certificates where trusted authorities can verify and revoke verification for digital assets.

I was thinking the same thing. Apple and Google can start (as tech companies) to add signatures to images. As long as private key doesn't leak we are good. And each manufacturer can have different cert issued by higher authority e.g. Google, so it can be revoked when it leaks. They could include digital camera manufacturers (sony, nicon, canon, ...) and define a standard. Signature can be a meta tag based on hash of the image.

This would limit authenticity to images taken by official software.

You can still just manipulate the official hardware to produce the image you desire, i.e. record a video that's projected onto a wall. And it'd be fairly easy to do with existing technology too.

Wouldn’t it end up being some form of nuisance proof? Introducing friction to create verification check points or examine and establish the chain of evidence?

But that will only matter for highly legal things or important things. Everything else will be too much of a bother to follow information hygiene.

It’s like we’ve introduced an information weed, whose only goal is to create content that matches our dopamine receptors. Maybe our instincts will shift to assuming any shiny or eye catching content is fruit of the weed? Designed to attract us?

back to the blockchain!

Of course. I already imagine an end-to-end hardware DRM pipeline where images can only be modified with the software made by "trusted" certified parties. Mandated by law and tied to your real ID, of course. Analog loophole can be dealt with later, first things first. /s

Reply on news.ycombinator.com